Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let us consider two spaces $\mathbb{X},\mathbb{Y}$. For simplicity we put $\mathbb{X} = \mathbb{Y} = \mathbb{R}$. On the product space $\mathbb{S} = \mathbb{X}\times \mathbb{Y}$ we consider a Markov process $S = (X,Y)$ with a transition kernel $T$ such that $$ T(B|s) = \mathsf P\{S_1\in B|S_0 = s\}. $$

For any fixed $y\in\mathbb Y$ we construct a Markov process $X^y$ on $\mathbb{X}$ such that its transition kernel $T^y$ is given by $$ T^y(A|x) = T(A\times \mathbb Y|(x,y)). $$

For a set $A\subset \mathbb X$ we are interested in the following probability $$ u(s) = \mathsf P\{S_n\in A\times \mathbb Y\text{ for all }n\geq 0|S_0 = s\}. $$

What I have is that for all $y\in \mathbb Y$ and $x\in\mathbb X$ holds $$ \mathsf P\{X^y_n\in A\text{ for all }n\geq 0|X_0 = x\} = 0. $$

Can I derive from here that $u(s)\equiv 0$ or if there is a counterexample?

It will be even helpful if $\mathbb{X},\mathbb{Y}$ are finite (I mean for the proof or a counterexample).

share|improve this question

1 Answer 1

up vote 2 down vote accepted

Let $\mathbb{X}=\mathbb{Y}=\mathbb{R}$. Suppose that $$ P(S_1\in B\mid S_0=(x,y)) = \delta_{(x-2^{-y},y+1)}(B). $$ (Here, $\delta_z$ will always denote the point mass at $z\in\mathbb{R}^n$.) In other words, the process $S$ is a (deterministic) Markov chain that goes from $(x,y)$ to $(x-2^{-y},y+1)$ with probability one. In this case, \begin{align*} P(X^y_1\in A\mid X^y_0=x) &= P(S_1\in A\times\mathbb{R} \mid S_0=(x,y))\\ &= \delta_{(x-2^{-y},y+1)}(A\times\mathbb{R})\\ &= \delta_{x-2^{-y}}(A). \end{align*} In other words, $X^y$ is a (deterministic) Markov chain that goes from $x$ to $x-2^{-y}$ with probability one.

Now let $A=(0,\infty)$. For all $x,y\in\mathbb{R}$, as soon as $n\ge x2^y$, we will have $X^y_n = x - n2^{-y} \le 0$, and hence, $X^y_n\notin A$. It follows that $$ P(X^y_n\in A\text{ for all $n\ge0$}\mid X_0=x) = 0. $$ On the other hand, if $S_0=(x,y)$, then $$ S_n = \bigg(x-2^{-y}\sum_{j=0}^{n-1}2^{-j},y+n\bigg). $$ Letting $n\to\infty$, we see that the first component of $S_n$ will be positive for all $n$ if and only if $x-2^{-y+1}\ge0$. Thus, any $s=(x,y)$ which satisfies $x\ge2^{-y+1}$ (for example, $s=(1,1)$), will also satisfy $$ P(S_n\in A\times\mathbb{R}\text{ for all $n\ge0$}\mid S_0=s) = 1. $$ Informally speaking, the process $(X_n^y,y)$ (where $y$ is fixed) is constructed by taking $S_n$ and just forcing its second coordinate back to $y$ after each step. The process $S_n$, on the other hand, is allowed to wander freely in the $y$-direction according to the original dynamics. Through this wandering, we are able to produce different asymptotics, even in the $x$-direction, for the two processes $(X_n^y,y)$ and $S_n$.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.